Every server analysis Interact produces lands on a single rating per machine: a grade from A+ at the top to F at the bottom, scored in transactions per Watt. The grade is the headline output of every report. The methodology behind it is worth setting out clearly, because the grading principle is what makes the analysis useful for procurement.

What the grade measures

The grade is a per-configuration efficiency rating. That phrase carries the work. Published efficiency benchmarks for server hardware are typically reported one configuration per make and model, picked by the manufacturer or by SPEC. That reference configuration is rarely the configuration sitting in the customer's rack. RAM differs. CPU SKU differs. Drive setup differs. The differences are not trivial; the same server model in two configurations can fall several grades apart in measured efficiency.

The Interact grade is calculated against the specific configuration on site, not the reference build. As far as we are aware, no other tool on the market does this at the per-configuration level.

What you give us, what you get back

Five inputs:

  • Number of CPUs
  • CPU model
  • Server make
  • Server model
  • Total RAM

Output: a per-server grade from A+ at the top to F at the bottom, reported in transactions per Watt, with the dataset and methodology underneath available for inspection on request.

Why a grade rather than a number

Procurement teams already work with grades. The energy label on a fridge, the efficiency rating on a building, the ESG rating on a fund. The instinct to read A through F as "fix the bad ones first" is well-trained. The grade is the same instruction at the rack level.

The number underneath, operations per Watt, is the dataset talking. It matters when comparing two candidate replacements head to head, or when modelling the energy bill against a refresh scenario. The grade is what tells a CIO or CFO where to start looking.

Why this is better than ENERGY STAR

ENERGY STAR is the US government's well-known efficiency mark, recognised across the industry and built into procurement codes worldwide. For servers, it is granted to machines that meet a single threshold derived from SPEC's SERT score. The mark is pass or fail. Either the server qualifies, or it does not.

Pass or fail works well for products with limited configurability. Fridges, dishwashers, monitors. The consumer does not choose how many compressors a fridge has; what gets tested is essentially what gets sold.

Servers are different. A server is sold more as a bill of materials than an off-the-shelf product. CPU count, CPU model, RAM capacity, RAM rank, drive setup, all configurable at the order. Each of those choices materially changes the energy efficiency of the resulting machine. The same server model in two configurations can land 40% apart on measured performance per Watt. An upgraded immediate-past-generation server can outperform the base configuration of the latest generation by more than a third.

Under pass or fail certification, all of that variance disappears into one binary mark. A server that scrapes through certification and a server that comfortably exceeds it look identical on the sticker. Servers configured for backup or resilience, which run at low power states most of the time, are often missing from the certified list entirely, because the certification does not address that operating profile.

A per-configuration grade carries the information the binary mark hides. The legislators currently building server efficiency standards, including Singapore's national framework and the EU's Ecodesign Directive work, are looking for exactly this kind of granular, configuration-aware rating. We have spent years building it, because the binary one was never going to be enough.

What customers do with it

Three things, repeatedly:

  • Concentrate on the bottom of the curve. Decommission, consolidate or replace the lowest grades first. Highest carbon and cost return per unit of action.
  • Plan refreshes around configurations, not models. A vendor-recommended like for like replacement often lands in a different grade than expected because the configuration shifts. The grade flags those before the purchase order leaves the building.
  • Stage a digital transformation against measured starting points. Scope 2 and 3 reductions, TCO reductions, facility consolidation, rack densification, and each is easier to plan against a graded estate than a flat list.

When the grade is the wrong tool

For two cases. If the question is about a workload running on a single server rather than a hardware decision, the grade alone is not the answer; workload profiling sits next to it. And for very small estates of single-digit servers, the grade is over-engineered; a conversation gets there faster.

For everything else, the grade is where every Interact analysis starts.